Reconstruction filter

Reconstruction filter is the filter applied to a sampled discrete signal to recover an approximation of the original continuous signal. In the ideal case specified by the Nyquist-Shannon sampling theorem, the reconstruction filter is the sinc function — the inverse Fourier transform of the ideal rectangular low-pass filter that isolates the baseband spectrum from its replicated aliases. The sinc filter has infinite support in the time domain, meaning that perfect reconstruction of any single point requires knowledge of all samples, extending infinitely into the past and future.

Real systems cannot implement the ideal sinc filter. The impulse response must be truncated, windowed, and approximated by finite-order filters. Common approximations include the zero-order hold (the simplest reconstruction, which outputs each sample value until the next sample arrives), the first-order hold (linear interpolation between samples), and higher-order filters using FIR or IIR architectures. Each approximation introduces a different error profile: zero-order hold produces staircase distortion; linear interpolation produces piecewise-linear artifacts; truncated sinc approximations introduce ringing and phase distortion.

The choice of reconstruction filter is a trade-off between computational cost, latency, and fidelity. The ideal sinc filter is non-causal and infinitely long, requiring lookahead that no real-time system can provide. Causal approximations sacrifice high-frequency accuracy for realizability. The Gibbs phenomenon — the overshoot near discontinuities in truncated Fourier series — appears in reconstruction as ringing at sharp edges, a visible artifact in image reconstruction and an audible pre-echo in audio upsampling.

In digital-to-analog conversion, the reconstruction filter is the final stage before the signal enters the analog domain. The quality of this filter determines the extent to which the digital representation actually corresponds to the continuous waveform it claims to represent. A poorly designed reconstruction filter can introduce aliases that were not present in the digital signal, or it can attenuate frequencies that should have been preserved, turning the promise of the sampling theorem into a practical failure.

The reconstruction filter is the forgotten half of the digital promise. The sampling theorem tells us that samples are sufficient; the reconstruction filter tells us that sufficiency is not the same as accessibility. The ideal reconstruction exists in the mathematics but not in the hardware. Every digital system lives in the gap between the theorem and the filter, and the quality of that life is determined by how well the approximation conceals the gap. We are not living in a digital world. We are living in a world of approximate reconstruction, and the approximation is never perfect.